I feel like this post is addressing an argument that isn’t the actual argument that MEV auction proponents (including myself) are making! In particular:

> But the funds “created” by an MEV auction don’t come from nowhere. Ultimately the money comes out of the pockets of users. Here’s why. Miner-extractable value currently goes to the miners. There is some amount of value that we need to pay to the miners to keep enough of them mining … one way or another, users of the system are paying those fees, whether in gas, inflation, or value lost to front-runners.

Of course the funds collected by the MEV auction are taken out of users’ pockets, and of course if we separate sequencing from mining then the miners will have less revenue. However, we’ve known all of this from the beginning! The benefits of MEV auctions are not about making new money come out of nowhere, they are explicitly about redirecting a money stream from miners (and in eth2 validators) to rollup sequencers.

This is done for two reasons:

1. MEV extraction is naturally a highly high-economies-of-scale and centralization-prone activity. It involves writing and constantly maintaining advanced software. It’s arguably even more high-fixed-cost than mining itself, and definitely more so than PoS. Separating out MEV extraction and auctioning it off essentially quarantines this centralization-prone revenue stream by assigning it to separate actors, reducing the risk that mining and later PoS will centralize. I made this argument before here: https://twitter.com/VitalikButerin/status/1247995681459011591
2. Currently, there are political constraints (eg. see responses here) that make any eth1 base-layer public goods funding difficult; basically, it’s hard to fund public goods in a way that’s credibly neutral. As a result, nearly all blockchains are heavily overinvesting in security (which we do have clean-and-simple taxes for) and underinvesting in other public goods. MEV auctions siphon off this revenue from layer 1 into layer 2, which does not have these political constraints (because it’s easier to coordinate escaping from a corrupted layer 2 than from a corrupted layer 1), and so can be used to fund(marginally higher value-per-dollar) non-security public goods.

Additionally (HT Ben Jones for reminding me of this), it’s not clear that MEV revenue in the hands of miners increases security. MEV revenue is txfee revenue, and it’s a long-time known fact that txfee revenue is more prone to encouraging 51% attacks and other malicious behavior than stable predictable block reward revenue. Basically, because of MEV’s highly volatile nature (and MEV increases if you are reorging the chain!) it creates lots of occasional opportunities where short-range chain reversions become profitable. MEV being captured by a sequencer has none of these problems.

>There is an ideal level of tax revenue to collect, balancing your revenue goals against the distortions caused by the tax. It’s very unlikely that the ideal tax size happens to coincide with the amount of MEV available on your chain. There are better and worse ways to distribute the tax burden across your population of users. It’s unlikely that taxing people in proportion to their use of front-runnable services is an efficient or fair distribution of the tax burden.

I think this ignores the fact that very severe informational and political constraints in public blockchains put a pretty low upper bound on how fair a tax system can be. For example, you can’t have proportional-to-value-sent transaction taxes, because there is an unlimited number of ways to obfuscate how much value a transaction is transferring (even rollups themselves would evade a transaction tax on the base chain!). At this point, any tax revenue going into a pool that is legitimately usable for anything but paying for security is a huge step up for the ecosystem.

> Centralizing MEV extraction

As mentioned above, centralizing MEV extraction is good because it quarantines a revenue stream that could otherwise drive centralization in other sectors. It is true that a centralized MEV environment can extract more MEV than a competitive one, but this ignores that a competitive MEV environment has other negative externalities, eg. it bloats the chain with a high rate of failed transactions. And there’s an upper bound on how harmful centralized MEV extraction can get, because users always have the ability to switch to application patterns that escape it.

> That still means that the current event ordering dictator can impose an extra ten minutes of delay on all of your transactions.

I *believe* that the technique being considered to prevent this is to have some kind of larger voting pool that can kick out a sequencer (potentially without compensation). In addition to preventing malicious censorship, this would also prevent the most harmful but low-value forms of MEV extraction.